ReviewCritical review of receptor modelling for particulate matter: A case study of India
Highlights
► Many studies in India have used multivariate statistical models and a few CMB. ► The results are diverse and not explained entirely by different sampling locations. ► Methodological issues are identified. ► Genuine overlaps in chemical composition between soil and road dust create problems. ► Recommendations are given for future studies.
Introduction
Air quality has been a cause of concern all over the world with the concentrations of criteria pollutants exceeding the standards at many places, particularly in developing countries. Particulate matter (PM) has been recognized as one of the key pollutants with a negative impact on human health, and a range of regulations have been introduced in order to control PM10 levels in urban areas with an increasing focus on PM2.5 control. However, in order to design effective programmes and strategies for reduction of PM concentration in the ambient air, it is necessary to have information about the sources and their respective contributions.
The term, source apportionment (SA) describes techniques used to quantify the contribution of different sources to atmospheric PM concentrations. There is a wide range of published literature on source apportionment using dispersion models and monitoring data (Laupsa et al., 2009; Colvile et al., 2003). However, in the Indian context, most of the source apportionment studies have been conducted using receptor models and hence, receptor models are the focus of this review. Receptor models form a subset of source apportionment techniques and apportion the pollutant concentrations based on the measured ambient air data and the knowledge about composition of the contributing sources (Henry et al., 1984). The key outputs are the percentage contributions of different sources to pollutant concentration. Such models are particularly helpful in cases where complete emissions inventories are not available (Hopke, 1991). Receptor models have been used for identification of sources and their respective contributions to airborne particulate matter across the world (Harrison et al., 1997; Kumar et al., 2001; Larsen and Baker, 2003; Begum et al., 2004; Lai et al., 2005; Song et al., 2006; Tsai and Chen, 2006; Chowdhury et al., 2007; Guo et al., 2009; Kong et al., 2010; Stone et al., 2010; Gu et al., 2011).
Receptor models can be divided into two broad categories: microscopic and chemical. Microscopic methods, including optical, scanning electron microscope (SEM) and automated SEM analyses are primarily based on the analysis of morphological features of many individual particles in the ambient air (Cooper and Watson, 1980). However, they are not very feasible for large-scale use since they do not produce quantitative results in most cases. Chemical methods, on the other hand, utilize the chemical composition of airborne particles for identification and apportionment of sources of PM in the atmosphere. A number of different models are included in this category such as enrichment factor analysis, times series analysis, Chemical Mass Balance (CMB) analysis, multivariate factor analysis (including Principal Component Analysis (PCA) and Positive Matrix Factorization (PMF)), UNMIX, species series analysis and Multilinear Engine (ME) analysis (Cooper and Watson, 1980; Henry et al., 1984; Hopke, 1991; Ramadan et al., 2003). Such methods use trace elements, elemental/organic carbon and organic molecular markers for identification of sources and over time have become popular for SA analyses.
Since PM is composed of both inorganic (trace metals, cations and anions) and organic species, a range of source markers are used in receptor modelling studies. Traditionally, most studies were carried out using inorganic trace elements like Fe, Zn, Pb, Cr, Al and Ni. However, since many of the trace elements are emitted from a range of sources (e.g., Zn is emitted from tyre wear as well refuse burning), it was difficult to apportion the PM to sources with a high degree of confidence. Further, with the removal of elements like Pb and Br from gasoline, there has been a need to develop and use new markers. In the last two decades, research has focused on the identification and development of organic molecular markers for SA since they can be characteristic of sources, thus reducing the source ambiguity, and creating markers for sources which are difficult to be apportioned solely on the basis of inorganic tracers, e.g., levoglucosan for biomass burning (Harrison et al., 1996, 2003; Schauer et al., 1996; Robinson et al., 2006).
The CMB method requires a priori knowledge of the composition of all sources contributing to the airborne pollution, but not their emission rates. The measured air quality is assumed to be a linear sum of the contributions of the known sources, whose contributions are summed over each different sampling period to give the best match to the concentrations of the many chemical species measured in the atmosphere. In more recent studies, organic “molecular markers” which may be only minor constituents of emissions are measured, as these help to discriminate between similar sources (e.g., gasoline and diesel engines).
There is a suite of multivariate statistical methods based upon factor analysis, of which PMF has been developed specifically for the purpose of source apportionment of air quality data, and is the most commonly applied. The method requires no a priori knowledge of source composition, but any information on source emissions characteristics is helpful in discriminating between similar sources. The method requires a substantial number (at least 50) of separate air samples and works best with a large dataset in which the number of samples far exceeds the number of analytical variables. A minimum variable to case ratio of 1:3 should be maintained in order to obtain accurate results (Thurston and Spengler, 1985). For a clearer distinction, it is better to have short sampling times so that overlap of multiple point source contributions to a given sample is minimised. The samples are analysed for the chemical constituents, and those constituents from the same source have the same temporal variation, and if unique to that source are perfectly correlated. Typically, however, a given chemical constituent will have multiple sources and the program is able to view correlations in a multidimensional space and can generate chemical profiles of “factors” with a unique temporal profile characteristic of a source. Past knowledge of source chemical profiles is used to assign factors to sources, and typically identification of six or seven different sources is a good outcome. Before PMF became widely adopted, PCA was widely used for the same purpose, but is less refined than PMF. Input data plays an important role in the final results, and care has to be taken to ensure that this is of good quality and where possible uncertainties can be assigned to individual analytes.
The key differences between CMB and the methods based upon multivariate statistics are summarised in Table 1. Studies have been conducted to compare results from different models (Larsen and Baker, 2003; Ramadan et al., 2003; Shrivastava et al., 2007; Bullock et al., 2008; Lee et al., 2008; Viana et al., 2008b; Yatkin and Bayram, 2008; Callén et al., 2009; Tauler et al., 2009). Multicollinearity can affect the model estimates, particularly in cases where different sources have similar signatures, although multivariate models help to reduce that problem substantially (Henry et al., 1984; Thurston and Lioy, 1987). It has been reported that in cases where two different sources have similar signatures, it becomes difficult to distinguish between them and neither CMB nor multivariate models can distinguish between sources with similar signatures when additional information (for e.g., meteorology data) is missing (Henry et al., 1984).
Hybrid models such as target transformation factor analysis (TTFA) and the constrained physical receptor model (COPREM) have been designed to combine the features of CMB and factor analysis models with the aim of maximizing the advantages while minimizing the limitations of each model (Wahlin, 2003; Viana et al., 2008a). The Multilinear Engine (ME) program also allows the use of source composition data to constrain the model.
Larsen and Baker (2003) compared three different multivariate techniques- UNMIX, PCA/MLR and PMF for SA of ambient polyaromatic hydrocarbons (PAHs) in Baltimore. Although they reported that PCA/MLR is unable to model extreme data effectively, they concluded that the overall source contributions compare well among the various models. They also reported that use of different techniques on the same data set could help in identification of missing sources, and increase the robustness of the results. Shrivastava et al. (2007) used PMF and CMB for source apportionment of organic carbon and found good correlation between individual profiles for CMB and factors identified by PMF but with systematic biases that were found to be within an acceptable range (a factor of two). Lee et al. (2008) compared the CMB and the PMF models and concluded that although both models identify similar sources, they apportion contributions of different sources differently. The authors suggested that a lack of local source profiles, omission of key sources or lack of suitable markers, and the different assumptions regarding aging of the source emissions as the possible causes for the different estimations. Viana et al. (2008b) compared PCA, CMB and PMF for identification of source contributions to PM10 in Spain. They reported overall consistency between the different models with high correlation in terms of source identification. However, they noted larger differences in terms of the percentage contribution of various sources. They suggested that a combined approach with the use of multivariate techniques for identification and interpretation of emissions sources and use of CMB for source contribution could help in increasing the robustness of the results. Earlier, Thurston and Lioy (1987) had also suggested a similar approach with the consecutive use of multivariate and chemical mass balance models to derive better results from receptor modelling studies. Similarly, Shi et al. (2011) tested a combined two-stage PCA/MLR- CMB model and found acceptable results using synthetic datasets with collinearity. They also concluded that maximum uncertainty is generally observed in case of highly collinear sources.
Callén et al. (2009) compared three different multivariate techniques- PCA-ACPS, UNMIX, PMF for source apportionment of PM10 and found that the different models showed high correlation between modelled and measured concentrations and PCA and PMF were able to identify more sources in comparison with UNMIX with good agreement. Tauler et al. (2009) compared four different multivariate models (PCA, PMF, Multivariate Curve Resolution by Alternating Least Squares, (MCR-ALS) and Weighted Alternating Least Squares (MCR-WALS)) and concluded that PMF and MCR-WALS identify sources and apportion the emissions to sources in a similar fashion. The weighted models (PMF and MRC-WALS) were found superior in robust and accurate factor identification.
Receptor models have been used for regulatory purposes since they were first used in Oregon, USA in the late 1970s (Gordon, 1988). However, there is a caveat regarding the degree of uncertainty associated with the results (Caselli et al., 2006).
Given the rapid rates of urbanization in Indian cities, air pollution is increasingly becoming a critical threat to the environment and to the quality of life among the urban population in India. Air quality has been a cause of concern in Indian cities with the concentrations of criteria pollutants exceeding health-based standards, and PM has been identified as one of the key public health concerns. High enrichment factors have been reported for various metals including Pb, Zn, Cu, Ni and Cr in a number of Indian cities, indicative of anthropogenic sources of heavy metals in particulate matter (Kulshrestha et al., 1995; Pandey et al., 1998; Negi et al., 2002; Rastogi and Sarin, 2009). Also, using SEM–EDX analysis, Srivastava et al. (2009b) reported that particles were primarily of anthropogenic origin irrespective of size range in polluted areas, e.g., traffic intersections. Although there has been an increased focus on PM emission control in recent years, the concentrations are still found to exceed the National Ambient Air Quality Standards (NAAQS) regularly.
The primary sources of air pollution in India have been identified as vehicular emissions, industrial emissions, coal combustion, biomass burning, road dust and refuse burning. There has been a rapid increase in motorization in India in the past years and this has led to an increasing contribution of the transport sector to air pollution in urban areas. Small-scale industries functioning within urban centres have been found to contribute to the air pollution problem.
Section snippets
Source apportionment and receptor modelling in India
There has not to our knowledge previously been a review of either aerosol source apportionment or receptor modelling work conducted in India. In this article, we seek to review existing knowledge and to make recommendations as to future directions. There is a growing body of literature on source apportionment of PM in India using receptor modelling (Table S1 in Supplementary Information). A majority of the SA studies have been conducted using multivariate methods; PCA being the most commonly
Discussion and conclusions
There have been many studies conducted in India using receptor modelling methods for source apportionment of particulate matter. India is a very large and diverse country, and unsurprisingly the studies have drawn widely differing conclusions. Even within individual Indian cities, different authors have come to widely varying conclusions over source attribution and apportionment, and this may to some extent be a result of using different sampling locations and seasons. Most studies have
Acknowledgement
Pallavi Pant gratefully acknowledges financial support from the University of Birmingham fund for collaboration with India.
References (112)
- et al.
The use of trajectory cluster analysis to examine the long-range transport of secondary inorganic aerosol in the UK
Atmospheric Environment
(2005) Concentration level, pattern and toxic potential of PAHs in traffic soil of Delhi, India
Journal of Hazardous Materials
(2009)- et al.
Source apportionment of atmospheric urban aerosol based on weekdays/weekend variability: evaluation of road re-suspended dust contribution
Atmospheric Environment
(2006) - et al.
Particle size distribution and its elemental composition in the ambient air of Delhi
Environment International
(2000) - et al.
Elemental composition and sources of air pollution in the city of Chandigarh, India, using EDXRF and PIXE techniques
Nuclear Instruments and Methods in Physics Research Section B: Beam Interactions with Materials and Atoms
(2000) Heavy metal levels and solid phase speciation in street dusts of Delhi, India
Environmental Pollution
(2003)- et al.
Assessment of heavy metal content in suspended particulate matter of coastal industrial town, Mithapur, Gujarat, India
Atmospheric Research
(2010) - et al.
Investigation of sources of atmospheric aerosol at urban and semi-urban areas in Bangladesh
Atmospheric Environment
(2004) - et al.
Composition and size distribution of particules emissions from a coal-fired power plant in India
Fuel
(2008) - et al.
Positive matrix factorization and trajectory modelling for source identification: a new look at Indian Ocean experiment ship observations
Atmospheric Environment
(2008)
Source apportionment of PM10 in six cities of northern China
Atmospheric Environment
Evaluation of the CMB and PMF models using organic molecular markers in fine particulate matter collected during the Pittsburgh air quality study
Atmospheric Environment
Comparison of receptor models for source apportionment of the PM10 in Zaragoza (Spain)
Chemosphere
Use of dispersion modelling to assess road-user exposure to PM2.5 and its source apportionment
Atmospheric Environment
Receptor modelling using positive matrix factorization, back trajectories and Radon-222
Atmospheric Environment
Metals associated with airborne particulate matter in road dust and tree bark collected in a megacity (Buenos Aires, Argentina)
Ecological Indicators
Identification of brake wear particles and derivation of a quantitative tracer for brake dust at a major road
Atmospheric Environment
Air quality impact assessment of NOx and PM due to diesel vehicles in India
Transportation Research Part D
Source apportionment of ambient particles: comparison of positive matrix factorization analysis applied to particle size distribution and chemical composition data
Atmospheric Environment
Receptor modelling of source apportionment of Hong Kong aerosols and the implication of urban and regional contribution
Atmospheric Environment
Chemical mass balance source apportionment of PM10 and TSP in residential and industrial sites of an urban region of Kolkata, India
Journal of Hazardous Materials
Comparative receptor modelling study of airborne particulate pollutants in Birmingham (United Kingdom), Coimbra (Portugal) and Lahore (Pakistan)
Atmospheric Environment
A study of trace metals and polycyclic aromatic hydrocarbons in the roadside environment
Atmospheric Environment
Review of receptor model fundamentals
Atmospheric Environment
Comparative receptor modelling study of TSP, PM2 and PM2−10 in Ho Chi Minh city
Atmospheric Environment
An introduction to receptor modelling
Chemometrics and Intelligent Laboratory Systems
Source apportionment of PM10 at residential and industrial sites of an urban region of Kolkata, India
Atmospheric Research
Elemental characterization and source identification of PM2.5 using multivariate analysis at the suburban site of north-east India
Atmospheric Research
Receptor modelling of PM2.5, PM10 and TSP in different seasons and long-range transport analysis at a coastal site of Tianjin, China
Science of the Total Environment
Metal concentration of PM2.5 and PM10 particles and seasonal variations in urban and rural environment of Agra, India
Science of the Total Environment
Source apportionment of suspended particulate matter at two traffic junctions in Mumbai, India
Atmospheric Environment
Receptor modeling of source contributions to atmospheric hydrocarbons in urban Kaohsiung, Taiwan
Atmospheric Environment
Source apportionment of PM2.5: comparing PMF and CMB results for four ambient monitoring sites in the southeastern United States
Atmospheric Environment
Source apportionment at mixed urban site in Indonesia using PMF
Atmospheric Environment
Source apportionment of particulate matter (PM2.5) in an urban area using dispersion, receptor and inverse modelling
Atmospheric Environment
Discrimination between anthropogenic (pollution) and lithogenic magnetic fraction in urban soils (Delhi, India) using environmental magnetism
Journal of Applied Geophysics
Trace elemental composition of atmospheric particulate at Bhilai in central-east India
Science of the Total Environment
Vapour pressure of ammonium chloride aerosol: effect of temperature and humidity
Atmospheric Environment
Comparison of positive matrix factorization and multilinear engine for the source apportionment of particulate pollutants
Chemometrics and Intelligent Laboratory Systems
Quantitative chemical composition and characteristics of aerosols over western India: one-year record of temporal variability
Atmospheric Environment
Emissions inventory of anthropogenic PM2.5 and PM10 in Delhi during Commonwealth Games 2010
Atmospheric Environment
Preliminary study of the sources of ambient air pollution in Serpong, Indonesia
Atmospheric Pollution Research
Chemical speciation of respirable suspended particulate matter during a major firework festival in India
Journal of Hazardous Materials
Source apportionment of airborne particulate matter using organic compounds as tracers
Atmospheric Environment
Preliminary chemical characterization of particle-phase organic compounds in New Delhi, India
Atmospheric Environment
Assessment of ambient air PM10 and PM2.5 and characterization of PM10 in the city of Kanpur, India
Atmospheric Environment
Characterization and source identification of polycyclic aromatic hydrocarbons (PAHs) in the urban environment of Delhi
Chemosphere
Estimated contributions and uncertainties of pca/mlr-cmb results: source apportionment for synthetic and ambient datasets
Atmospheric Environment
Metallic species in ambient particulate matter at rural and urban location of Delhi
Journal of Hazardous Materials
Sources of organic aerosol: positive matrix factorization of molecular marker data and comparison of results from different source apportionment models
Atmospheric Environment
Cited by (308)
Sewage treatment plant dust: An emerging concern for heavy metals-induced health risks in urban area
2024, Science of the Total EnvironmentSeasonal variations of metals and metalloids in atmospheric particulate matter (PM2.5) in the urban megacity Hanoi
2024, Atmospheric Pollution ResearchSource specific health risks of size-segregated particulate bound metals in an urban environment over northern India
2023, Atmospheric Environment
- 1
Also at: Centre of Excellence in Environmental Studies, King Abdulaziz University, PO Box 80203, Jeddah (21589) Saudi Arabia.